diff --git a/nanobot/agent/runner.py b/nanobot/agent/runner.py
index 6b854a9a8..83438dbef 100644
--- a/nanobot/agent/runner.py
+++ b/nanobot/agent/runner.py
@@ -69,6 +69,8 @@ _COMPACTABLE_TOOLS = frozenset({
     "read_file", "exec", "grep", "find_files",
     "web_search", "web_fetch", "list_dir", "list_exec_sessions",
 })
+# read_file is the recovery path for persisted results; exempting it prevents persist->read->persist loops.
+_TOOL_RESULT_OFFLOAD_EXEMPT_TOOLS = frozenset({"read_file"})
 _BACKFILL_CONTENT = "[Tool result unavailable — call was interrupted or lost]"
 
 # Backward-compatible module attribute for tests/extensions that monkeypatch
@@ -1114,6 +1116,9 @@ class AgentRunner:
         result: Any,
     ) -> Any:
         result = ensure_nonempty_tool_result(tool_name, result)
+        if tool_name in _TOOL_RESULT_OFFLOAD_EXEMPT_TOOLS:
+            # Exempt tools bound their own output; skip generic offload and truncation.
+            return result
         try:
             content = maybe_persist_tool_result(
                 spec.workspace,
diff --git a/tests/agent/test_runner_persistence.py b/tests/agent/test_runner_persistence.py
index d2bcfa9d4..3c9431751 100644
--- a/tests/agent/test_runner_persistence.py
+++ b/tests/agent/test_runner_persistence.py
@@ -123,6 +123,54 @@ def test_persist_tool_result_logs_cleanup_failures(monkeypatch, tmp_path):
 
     assert "[tool output persisted]" in persisted
     assert warnings and "Failed to clean stale tool result buckets" in warnings[0]
+
+
+async def test_read_file_result_is_not_offloaded(tmp_path):
+    """read_file must not trigger generic offloading (prevents persist->read->persist loops)."""
+    from nanobot.agent.runner import AgentRunner, AgentRunSpec
+
+    provider = MagicMock()
+    captured_second_call: list[dict] = []
+    call_count = {"n": 0}
+
+    async def chat_with_retry(*, messages, **kwargs):
+        call_count["n"] += 1
+        if call_count["n"] == 1:
+            return LLMResponse(
+                content="reading",
+                tool_calls=[ToolCallRequest(id="call_rf", name="read_file", arguments={"path": "big.txt"})],
+                usage={"prompt_tokens": 5, "completion_tokens": 3},
+            )
+        captured_second_call[:] = messages
+        return LLMResponse(content="done", tool_calls=[], usage={})
+
+    provider.chat_with_retry = chat_with_retry
+    tools = MagicMock()
+    tools.get_definitions.return_value = []
+    tools.execute = AsyncMock(return_value="x" * 20_000)
+
+    runner = AgentRunner(provider)
+    result = await runner.run(AgentRunSpec(
+        initial_messages=[{"role": "user", "content": "read big file"}],
+        tools=tools,
+        model="test-model",
+        max_iterations=2,
+        workspace=tmp_path,
+        session_key="test:runner",
+        max_tool_result_chars=2048,
+    ))
+
+    assert result.final_content == "done"
+    tool_message = next(msg for msg in captured_second_call if msg.get("role") == "tool")
+    # read_file result must NOT be offloaded to a file
+    assert "[tool output persisted]" not in tool_message["content"]
+    # read_file manages its own size; generic truncation must NOT apply
+    assert len(tool_message["content"]) == 20_000
+    # no file should have been written for this read_file call
+    offload_dir = tmp_path / ".nanobot" / "tool-results"
+    assert not any(offload_dir.rglob("call_rf.txt")) if offload_dir.exists() else True
+
+
 async def test_runner_keeps_going_when_tool_result_persistence_fails():
     from nanobot.agent.runner import AgentRunSpec, AgentRunner